Recovery-Oriented Computing: Main Techniques of Building Multitier Dependability

Frequent freezes and crashes on current systems bring tremendously heavy loads to the system administration, directly resulting in an undesirable increase on the total cost of ownership (TCO). Obviously, it is time to broaden the long lasting performance-dominated research focus, which has neglected other aspects of computing such as dependability, availability and stability. Deeming that software bugs, hardware faults and operator errors are facts to be coped with, not problems to be solved, Recovery Oriented Computing (ROC) concentrates on building systems that recover fast when a fault does occur, instead of aiming for systems that never fail. To reach high dependability in the Internet service environment, ROC implements two building blocks for recovery, microreboot and system-level undo, which have proven effective in handling failures. A suitable benchmarking method is developed to quantify the impact of these effects on the system dependability.